The model is trained to extract answers to questions about covid19, but I also need to rank 100 covid19 papers on relevance to the question/search term.
hey @MSJohannessen, it sounds like you’re looking for a retriever-reader architecture - for that i’d suggest taking a look at haystack (built on top of transformers).
you can find a covid-19 example along the lines you’re talking about here: DataMuni: Building A Faster & Accurate Search Engine with Transformers & Haystack
in general, you can use squad2 models as baselines for the reader, but you’ll probably get better performance by fine-tuning them on your corpus
2 Likes